User:Phol ende wodan/DungeonOverhaul/Jonadab

This is a set of comments I drafted in response to Jonadab's dungeon overhaul proposal when he made it public in the summer of 2017. The comments were made with only cursory knowledge of NetHack Fourk, and what had been implemented since the proposal was written.

Starting Assumptions

* Stairs (and ladders, etc.), trapdoors, and holes will have (stored at
  generation) a specific destination, enough info to nail down branch, level,
  and coordinates.  Thus it will be possible to have stairs in more than one
  location on a level, which lead to different places -- either to different
  coordinates on the same destination level, or to different levels in
  different branches.

This is very interesting. I envision small one-off levels reachable by a stair - randomly generated, not true branches, but inaccessible any other way than by finding a random extra stair to it.

* It's good for non-randomly-generated levels to have multiple variants.  This
  enhances game-to-game variety and thus replay value, even if only by a
  little.  However, it's even better for a special level to have significant
  random components within it, hopefully something more impactful than a maze
  around the edge.

Multiple variants are perhaps even less important than good randomized design of a single variant.

Early Dungeon

The levels with the branch stairs to enter the Mines and Sokoban can both occur
at a range of depths, so they can both be made bones-eligible simply by removing
the extra set of stairs and saving a normal non-special bones level, since the
extra set of stairs is quite literally the only special thing about the level.
Remove the stairs, and ipso facto the resulting bones level becomes a normal
bones level.  Bones will still never be loaded at the depth where the branch
stairs are found, but some other player will have the branch entrance at a
different depth and thus be able to load the bones as a normal level at their
original depth.

Agreed, though it might be a bit odd if, for example, a player dies on one of the upstairs on the level with the Sokoban stairs, and those get removed. Should probably always choose the branch stairs to remove instead of picking randomly.

Regular filler levels in the early Dungeons of Doom will be room and corridor
levels, much as before, possibly with a small number of minor enhancements (such
as the occasional non-rectangular room).

Hmm, how about levels that are entirely rooms (every room is adjacent to only other rooms), can get from room to room by doors or secret doors. Same number of objects placed. Special rooms work the same. Vaults are 2x2 rooms that have no doors; either this is the only possible room with no doors, or other rooms can generate without doors as long as one can still get from the upstairs to the downstairs. Closets either don't exist or exist through some clever placement algorithm that allows them to stick out the side of a room. The per-room frequency of altars, sinks, fountains. traps etc is decreased, so the overall frequency of these on a level stays the same. Maybe slightly more objects generate because the level has more space and it's harder to explore the whole level.

The main change for the Big Room is that it should be guaranteed NOT to occur
before or on the same level with the entrance to Sokoban.  (The rationale here
is that players should have the option to complete Sokoban early, before
entering the Mines, and having the Sokoban entrance be in the Big Room makes
that difficult, especially for new players.  More experienced players should be
finding their challenges somewhat deeper into the dungeon or may choose to skip
Sokoban altogether.)

I propose that the effective depth of each Sokoban level should be the same as the level where the stairs to it are. 3.4.3 makes it too easy by decreasing the difficulty as you progress, 3.6.0 makes it too hard by increasing the difficulty (which doesn't scale with finding altars or better gear like descending the normal dungeon would). Keeping it static makes it still pursuable before doing the Mines, but doesn't make monsters too wimpy, especially in the zoo.

A possible addition: increase the difficulty of the final level so that the zoo is a bit more challenging.

I thought about adding more Big Room variants but concluded that it's
unnecessary, due to the fact that more than half the time it's not generated at
all anyhow; thus, when the Big Room is generated, it naturally feels like
variety.  Combined with the fact that there are already several versions of it,
I decided no further versions are needed.

Big Room could be an early trial ground for "applying random generation to large parts of levels". A few of the variants in 3.6.0 do a little of this, but I think nothing more complex than mazes.

Middle Dungeon

As before, in any given game, there shall always be one Tribute level in the
middle dungeon; however, the Rogue level show now be only one possibility.  Each
possible tribute level shall be designed to look and feel like one roguelike
game (other than NetHack itself), but the player shall only encounter one of
these tributes per game.  The various possible tribute levels are discussed
later.

Although it would be really cool, it is almost certainly not worth the amount of
effort that would be required to introduce a good Brogue tribute level, more's
the pity.

I don't think it's worth the effort of implementing different roguelikes' look-and-feel. NetHack is directly descended from Rogue, so it's at least somewhat justified to have a direct tribute to it. However, other roguelikes (besides forks and descendants of NetHack) have little to no substantial relation to the game. It's also annoying enough, as a new player, figuring out what the Rogue level is, and how it works. Personally, when I first discovered it, I got very confused trying to open the doorways with no door.

The benefits of adding tributes to other games are very, very small compared to the added complexity to the developers and the added complexity to the player. In the existing Rogue level, there's not even any gameplay benefit, apart from the meager bones pile. There's not much reason to add more gameplay benefits here, either.

I'd probably even go further and say the Rogue tribute level shouldn't be guaranteed to appear. I don't see any reason why it's important to have in every game, apart from the fact that it's a tribute. Maybe only a 30% chance of generating it or something.

Medusa's Island is to have a larger number of variants (instead of only two, at
least four) and is to occur 3-5 levels above the terminus of the Dungeons of
Doom (i.e., there should be 2-4 levels between Medusa and the final level at the
bottom).  I thought about removing the no-teleport restriction from Medusa's
Island either entirely or removing the restriction only when Medusa has been
killed but concluded that having a non-teleport water level between the upper
and lower dungeons is strategically significant and should be retained.  Players
who dig past Medusa should need to find a way to traverse it (typically,
levitation; or levelport; or scrolls of earth; or whatever) before they can
return upstairs e.g. to get the stuff they left in their stash or to revisit
Minetown.  This makes bypassing the level by digging past it a tradeoff that has
a downside.  So Medusa's Island stays no-teleport, even when Medusa has been
killed, the same as in 3.4.

I think you get interesting strategy either way. However, 2-4 guaranteed levels between Medusa and the terminus seems long, perhaps because I'm envisioning it the same way as such levels are now in vanilla - usually mazes. It's really cool when you get a room-and-corridor level with a shop or a temple under Medusa, though.

Alternatively, make Medusa's level undiggable floor (it fills in with water regardless of where you dig, perhaps?) and remove the teleport restriction after she is killed. I think I like it better the way it is, though.

The levels between Medusa and the terminus are to be mazes and are to have item
generation and monster generation probabilities that are different than the rest
of the Dungeons of Doom and also different from Gehennom.  I am open to the
possibility that some of these mazes may have passage widths or corridor widths
of more than 1 tile.

Required maze levels (on the main branch of the dungeon) that have nothing special in them are a Bad Thing in my book, for several reasons:

1. The more "open" a level is (some special levels, plain room-and-corridor levels, but especially cavernous levels), the faster a player can traverse it. Mazes are about the most un-open level possible. This makes them long and boring.
2. Mazes frustrate the player because they tend to pick out an area of the map as a short-term goal, but the maze often winds in a different direction and it turns out that that area actually takes quite a long time to get to. Even if you dig down to locate the downstairs, it can be quite annoying to find a path between the stairs.
3. Perfect mazes have many of their own problems, like making it very easy to pass by a tiny side corridor that ends up containing the downstairs, creating lots of dead-ends and branches of dead-ends, or creating the very long paths between close points mentioned above.
4. Spellcasters, specifically high-level divination spellcasters and particularly Wizards, enjoy a very large and I would say unbalanced advantage over characters who can't cast these spells on maze levels. Magic mapping instantly shows you the stairs, allows you to chart a path, and lets you determine how many walls you would need to break through to make a better path. Detect treasure shows you all the loot you can get on the level, and lets you determine whether it's worth going after and how long it'll take you to get it. This can be done in two actions the moment you arrive on the level. Players who can't do this have to stumble around the level and may miss valuable loot. Contrast the standard room-and-corridor levels: even without spells, it's relatively easy to find all the rooms, all the dungeon features, and all the objects in the rooms, and it can be done fairly fast.
5. People with easy access to digging enjoy a similar though smaller advantage over those who don't have it. For the aforementioned magic mapper casters, it's relatively easy to optimize the number of times you zap or cast digging and conserve your resources. Likewise, characters that wield a digging tool as their main weapon can just bust through as many walls as they want. This isn't a criticism of those benefits to those roles as much as a criticism that roles without those benefits are either compelled to traverse the mazes or lug around a digging tool and frequently have to switch it out of hand for a primary weapon.
6. There's nothing to break the monotony except for monsters, traps, and the occasional item. There are no special features except the stairs. For later maze levels that contain a demon lair or the Wizard's Tower, there's either 0 or 1 possible points of entry, so they're not really that interesting either since they don't really interact with the maze. The Fake Wizard's Towers are not quite as bad, since they challenge the player to cross the moat and dig their way in. One thing that would be interesting is to randomize the location of the fake towers on the level.

These complaints are more aimed at the Gehennom-of-endless-repetitive-dreariness mazes, but most of it still applies to possible maze levels between Medusa and the terminus.

Late Dungeon

The mazes are gone; filler levels in Upper and Lower Gehennom will be generated
using a new algorithm, based on the new Gehennom cavern generator (per ais523),
with some adjustments.  There can now be lakes of water and/or lava, with
probabilistic bias toward water at earlier levels and lava deeper in; the
probability will also be slanted toward larger quantities at deeper levels.  In
some cases it may be necessary for the player to cross water or lava, but
*usually* it will be possible to go around.  The probability numbers are to be
tweaked so that levels with no water or lava occur sometimes, but levels with
multiple regions of water or lava also occur.  Lava should generally be lit, but
water and floor areas should usually be dark, and corridors in Gehennom should
always be generated dark.

I like this. Most variants tend towards a cavernous Gehennom, and they're on the right track. I don't quite like Un's "gnomish mines except with lava instead of rock" thing, and I don't like Grunt's "gnomish mines except with random lava pools and rivers and rock instead of walls" thing.

I think the ais caverns are a bit too plain and Big Room-style.

I can't think of a good set of alternate versions of the Vlad's Tower levels
that preserves the general feel of the branch and yet is still genuninely
different, on a level-by-level basis, from the existing versions.  Unless
someone else can come up with such a set, I propose that Vlad's Tower be left as
an exception to the "special levels should have alternate versions" rule.  Think
of it as the exception that proves the rule.  The loot in Vlad's Tower is also
basically unchanged.

Proposed top of Vlad's Tower: let it contain an open area where Vlad sits, rather than a corridor where you can bounce rays back at him and he can't do anything about it. How about just a big square with the three closets on each edge, with Vlad and his throne in the middle. For bonus points, randomize which six towers contain the vampires on chests and the one with the stairs. (3.6.0's lead should be followed here and two chests should contain 4d2 candles.)

With the open area to fight in, Vlad's AI tries to move him to places you can't hit him with a ray twice, and he will stay out of melee range if possible. (Grunt gives him a polearm to battle with; this would work well here.)

The Fire Pits are to be fire themed and will be the easier of the two paths, but
less rewarding.  The filler levels in the Fire Pits will be more than half lava
and be inhabited mostly by fire-resistant monsters, many of which have fire
attacks (some of which, hopefully, are only partially resistable with fire
resistance; the "wand balance" proposal would help significantly here).  To make
the lava even more interesting, at least one monster will be given an attack
that takes advantage of it.  Perhaps Salamanders will have a drown-in-lava
attack similar to the eels' drown-in-water attack.  Two demons will have lairs
here, but these lairs are redesigned to fit the fire/pit theme.  The first demon
lair in the Fire Pits is either Asmodeus or Baalzebub (equal chance), and the
second is either Yeenoghu or Dispater (equal chance).

The Swamp of Death will be swamp themed and will be more dangerous, but also
more rewarding, than the Fire Pits.  There will be more monsters and more
dangerous monsters here than in the Fire Pits.  Two demons have lairs here: the
first is Juiblex, and the second is Demogorgon, whose lair contains a guaranteed
magic lamp.

No love for Geryon? :(

Also, is there anything else making this "more rewarding" besides the magic lamp?

Lower Gehennom consists of a mixture of filler levels (several of which contain
embedded features: this is where the Wizard's Tower and fake wizard towers live)
and special levels, chiefly the Black Market.  The last level in Lower Gehennom
appears like an ordinary (new ais-caverns style) filler level, but instead of
being generated with a down stair it contains the Vibrating Square, which works
just about the same as in 3.4.3.

Vibrating Square should follow 3.6.0 rules and be detectable and visible as a trap.

Moloch's Sanctum will dispense with most of the riff-raff and make the High
Priest himself more dangerous.  My current thinking for this is a special attack
that turns random tiles in your general vicinity to lava, as well as a melee
attack that does shock damage (which is partially resistable but can still
destroy rings and maybe wands) and the destroy armor monster spell.  Salamanders
with their new drowning attack, to make the lava even more dangerous, can be
considered if additional difficulty is wanted, but that may not be necessary.

Here's a great way to buff the high priest without even needing to buff his attacks: give him a shield of reflection instead of a small shield. More powerful attacks don't matter if most characters can just wand-of-death him on sight. Also see later notes on him.

To dispense with the riffraff, consider nerfing summon insects. A swarm of priests of Moloch is a good obstacle; a swarm of ants is just annoying at that point. Is there a plan to remove the graveyard? Also see later notes on summon insects (permit it everywhere but the Sanctum and the Astral Plane) and riffraff disposal.

Gnomish Mines

Filler levels in the Gnomish Mines are to be very similar to version 3.4.  The
first level of the Mines is to be lit, guaranteed, but subsequent levels have
the same lighting odds as in version 3.4.

Alternatively, if the first level of the Mines is unlit, guarantee a lighting tool (not a magic lamp) a couple steps away from the stairs. This is independent of the normal tool that generates in the Mines.

Rather than having Gnomes carry light sources, as so many variants do, I propose
that (non-undead) dwarvish monsters in the Mines sometimes carry one, usually a
candle, occasionally a brass lantern.  The flavor explanation for this is that
while Gnomes do not need light for anything (they live exclusively underground
and have a magical affinity with the earth that lets them feel things like
minerals at a distance), dwarves do need it for some things.  Dwarves can see
better underground than a human (because, infravision), well enough to get
around and to defend themselves, but they need light e.g., to appraise gem
qualities.  This is why they carry light sources in the Mines in particular.
Perhaps they buy them in Minetown.  The real (play balance) reason for giving
light sources to dwarves, not Gnomes, is that dwarves are fewer in number than
Gnomes, so it's a less drastic change.  This is also the reason for only giving
them light sources sometimes, and only in the Mines, and usually just a single
candle.  This proposal is not strictly required in order to ensure enough
candles: another source of candles is added later.  Note too that NetHack4
currently gives stacks of candles to Gnomes, so for NetHack4 this actually
brings us back closer to vanilla.

The Lower Mines should be completely unlit (Mines' End special areas excepted), with the player expected to find light sources in Minetown or have otherwise obtained them. To balance this, Upper Mines levels now have only a 1/3, 1/4 or 1/5 chance of being unlit (not sure which is best balanced). I feel like this will add to the player's perception of the Lower Mines as a more dangerous area, as well as provide a justified increase in difficulty when descending below Minetown.

Minetown in principle is to be roughly the same as in version 3.4, with a few
tweaks.  The randomly-generated portions of Frontier Town, Town Square, Alley
Town, College Town, and Bazaar Town are now cavern-like, rather than being
composed of rooms and corridors.  Additionally, all doors in Minetown are
well-oiled, so you will not hear them open from outside line-of-sight.  Due to
these two changes, players arriving via the stairs may not always immediately
know that they have reached the Minetown level.

Very good ideas.

Perhaps a more controversial change, I propose that one general store be
guaranteed to be generated in Frontier Town, Town Square, and Bazaar Town.  This
is intended to ensure (barring bones -- more on that in a moment) that the
player always has the option to do price-checking in a reasonably convenient
manner that does not require credit cloning.  Just needing to drag everything to
the level is enough hassle.  The reason for excluding College Town from this
list is simply that the guaranteed book shop allows price-checking scrolls,
which is adequate to bootstrap the identification process.  (The other three
Minetown variants already guarantee the general store, so no change is required
there.)

Guaranteed price-id is a good idea; however, College Town should probably get a general store anyways, even a tiny one.

If Orcish Town were to be included as well, it could contain a guaranteed general store (which has been ransacked of anything of value, of course), and the shopkeeper has barricaded himself in with boulders. This is more of a 3.6.0 improvement, but I think it's pretty well balanced to round out the complaints about Orcish Town.

Early Dungeons of Doom

Falling rock traps should not be generated in the main dungeon until dungeon
level four.  This gives both the player and their starting pet a chance to kill
enough monsters to level up a couple of times and thus have enough HP to not die
instantly.  (In other branches, falling rock traps can be generated at any
dungeon level, as before.  Currently this is only relevant for the first Mines
level.)

Falling rock traps should still be generated at level 1, but their damage is capped at half the monster's maxHP, rounded up. Something else to remove YAAD-causing traps from the early levels is to prevent magic traps from generating till level 4 or 5.

The design of the regular room-and-corridor levels could be improved in a number
of additional ways, but these changes should be adquate for the time being.
More can be done in some later version, but honestly at this time I feel that it
is more important to focus most of the effort on other areas of the game,
particularly Gehennom.

They are good the way they are, but why not put the ideas out there anyway?

One mostly cosmetic change for variety: "corridors" in these levels or on some of these levels aren't corridor tiles cut out of rock, they're floor tiles bounded by walls. Rock moles and the like will still dig corridor tiles through rock.

Delphi is fine as it stands, except that the Oracle should charge less.
(Players reaching this depth in the dungeon should be able to afford at least
one major consultation, if they have picked up all the gold so far.)  Also, the
Oracle's repertoire of consultations should be expanded somewhat; the NH4 bug
tracker has a ticket for this.

It's not clear what "... if they have picked up all the gold so far" means, does this include starting gold? A Healer will probably be able to buy at least 2 consultations under vanilla prices, a Tourist may as well. In any case, yes, I propose that the cost of a major consultation should be 400 or 500 zorkmids, perhaps increasing by a certain amount with each new major consultation. Major consultations should be more accessible to unspoiled players.

Sokoban

There are several reasons why I feel more variants of the Sokoban levels would
be a Good Thing:

Randomly generating Sokoban is all the rage these days, and it's closer to completion than it was when this document was first written. I'm holding out for the game to be able to fully generate four levels of increasing difficulty within the game and not relying on an external program. I'd also take the option that allows generating a lot of .des files for Sokoban and letting the local player or server admin change them out as desired. Either one adds randomly generated content, eliminates "sameness", and eliminates the possibility of spoilers.

The fourth level of Sokoban will probably have only three versions (bag, amulet,
and ring), unless someone can think of a fourth prize that matches well
(balance-wise) against the other three, in terms of being preferred by a similar
percentage of the player base.  (The other option would be to maintain an even
number of variants for Soko4 and place the amulet in half of them and the bag in
the other half, leaving out the ring.)

I disagree with the idea that Sokoban should generate only one of three random prizes. One of the most frustrating things Sokoban can do to a game is to give you a second bag of holding or amulet of reflection when you really wanted something else, and makes all that pushing boulders around (or traversing Nabokos) seem pointless. Several variants fix this by generating all possible prizes, but only letting the player take one, and that's how I think Sokoban should work in general.

The new possible Sokoban prize that I propose is a ring of polymorph control.  I
feel that this provides a character survival benefit intermediate in value
between the bag of holding and the amulet of reflection; additionally, I feel
that it provides a fun/convenience benefit intermediate in value between the
amulet and the bag.  Thus, on both scales it is a middle ground.  Like both of
the others, it is an item that can be obtained elsewhere but can often be
difficult to find in the early game.  It is not actually needed, but it can be
really handy and in particular could be handy below Minetown (where polytraps
can occur with some frequency) but is certainly not required.

Ring of polymorph control is an interesting third-prize substitute for the cloak of magic resistance some variants add as a third prize. It's better versus polymorph traps than the cloak, but won't be as good later in the game if the player still hasn't found another source of magic resistance and is now facing spellcasters. Still, that problem doesn't go away in vanilla, so I think a ring of polymorph control as one prize works.

Additionally, the luck penalties for breaking the Sokoban rules shall only apply
if the character is NOT carrying a luckstone.  Any player who has been to Mine's
End and picked up the luckstone there should be able to breeze through Sokoban
using a pick-axe and the two provided scrolls of earth, with no luck penalty.
(A player who happens upon a luckstone on the floor before reaching the Sokoban
entrance can do the same thing.)  Additionally, once a given Sokoban level has
been completed, its luck penalties and other Sokoban-specific restrictions
(e.g., against rolling boulders diagonally) shall be lifted.

Disagree about the luckstone. First, they're not all that uncommon in the early dungeon or upper Mines. This could be made into a tighter constraint by using the achievement system instead of merely carrying a luckstone; i.e. you must have actually been to Mines' End and picked up the luckstone in order to remove this penalty.

However, I more generally disagree with the entire idea that completing Mines' End gives one a free pass to essentially cheat one's way through Sokoban. The whole point here is that the player must work their way through one of two branches, plus the zoo, to claim the reward; making the only remaining obstacle the zoo cheapens the prize. Additionally, I feel that it will encourage players to finish the mines before even trying Sokoban or Nabokos.

Removing the luck penalty after all pits/holes are filled is implemented in 3.6.0 and is a good change.

Third, as discussed elsewhere in this proposal, the average number of wands and
rings in the Mines will be very slightly increased, so that players who choose
to forego Sokoban (possibly completing Nabokos instead) will not be missing as
large a percentage of the available wands and rings in the early game.  I
propose that each filler level in the Mines has a 50% chance of an additional
random ring and, independently, a 50% chance of an additional random wand.  (It
is understood that players who do both Sokoban and the Mines will thus get a few
more rings and wands in the early game than was typical in version 3.4; but I do
not believe this difference will be significantly unbalancing, since rings and
wands also occur as death drops anyhow and become more common later in the
game.)

But keep the copious food limited to Sokoban. For certain characters, going after Sokoban for its food is a strategic decision.

Finally, the statue of Perseus is now guaranteed to contain both a shield of
reflection and a bag (usually a plain sack or in one variant an oilskin sack),
so players who press on can be assured of finding both these items there.  The
amulet of reflection or the bag of holding from Sokoban would be nicer, and
could be obtained without going so deep, but it is possible to live without.  In
3.4.3, if you skip Sokoban, you risked facing the Castle with no bag and/or no
reflection.  This is now no longer the case.

Guaranteeing shield of reflection is backed up in the myth and works well, I think, since shields are probably the weakest reflection source. Bag is also fine, if it can't be immediately informally identified, which it can't really if it's an oilskin bag or a sack.

All four Sokoban levels are to be made bones-eligible.  However, the bones
option shall default to 1, which means "only load bones on normal levels" (where
"normal" levels are mostly the levels where bones were possible in NetHack
3.4.3).  If the bones option is set to 0, that means never load bones at all,
and setting it to 2 means they can be loaded (very nearly) anywhere.  Only
players who have deliberately set the bones option to 2 can ever see Sokoban
bones.  By default, Sokoban bones will not be loaded.

I don't really like how bones:2 allows everything to be bones. Personally, I think the game loses something if Minetown were bones-ineligible, and I'd love to have bones on just about every special level, but I'd hate Sokoban bones because the dead character would probably have messed it up. Under this system there is no way to say I want to have Minetown bones but not Sokoban, and making it a compound option to specify which levels should be ineligible would reveal spoilers. I think Sokoban is best kept bones-ineligible.

Finally, every version of every Sokoban level is to be given a name, which is to
show up in #overview by default, once the level has been visited, unless the
player renames it (using #annotate).  This minor change has no impact within the
game itself but is intended to facilitate conversations about the game.
Currently we find ourselves referring to things like "Sokoban level 2a", which
indeed is slightly ambiguous since not all spoilers present the levels in the
same order.  (In particular, Steelypips and the wiki are at odds about this
nomenclature.)  Giving them all official names does away with that.

For full random generation of Sokoban, this doesn't work.

Nabokos

The five levels of Nabokos are intended to be more difficult than Sokoban or
even than normal main-dungeon levels at a similar depth, but not so difficult as
to be unreasonable, especially for a player who has already tackled the Mines.

This would be a good place to move the 3.6.0 Sokoban level depth increase; i.e. if the stairs to Nabokos are on level 7, the first level of Nabokos has an effective depth of 8, the second has a depth of 9, and so on.

Each level of Nabokos is composed of about 8-10 rooms packed close together and
joined by very short (1-4 tile long) corridors, making the entire level just a
bit larger than a fourth-level Sokoban level, with similar amounts of floor
space.  Each room is populated, initially at level generation time, with 2-5
monsters.  There are no special rooms in Nabokos, no vaults, and mineralization
is lighter (fewer buried coins and gems) than in the main dungeon.

Nabokos is intended to be entirely unrewarding. It should have no mineralization, and perhaps undiggable walls.

Nabokos does contain traps.  Each level has 3d3 traps placed randomly.  These
will not be statue traps (because there are no items, on the floor, including
statues) and will also not be rolling boulder traps (because there are no items
on the floor, including boulders), but there can be any other type of trap
appropriate for the level depth.  Doors may also be trapped per the usual
probabilities.  There are no chest or box traps simply because there are no
items on the floor (containers would count as items), and as noted stairs are
the only dungeon furniture.

Note effective level depth increase would allow harder traps - a good thing.

The monster generation rate in Nabokos is the same as in the main dungeon, in
number of monsters per dungeon level per unit time.

Also remove death-drops in Nabokos, or greatly reduce the chance of one; otherwise people will still try to clear Nabokos for them. Each level will have 16-50 monsters at generation, averaging about 5 or 6 death drops. These provide a rational incentive to clear Nabokos after doing Sokoban, which should not exist. They might even hang out in Nabokos afterwards to farm more monsters since they don't have to walk all the way across the screen to where something else was generated.

Lower Dungeons of Doom

Dragon halls, if we don't wish to reserve them for Gehennom, seem like an
obvious enhancement and could feature (sleeping) Great Dragons, normal adult
dragons (and possibly baby dragons, maybe even dragon eggs) of the same color,
and thematic loot.  Great Dragons are a step up from normal adult dragons.
Besides having higher stats and so being harder to kill, they also have more
powerful attacks, possibly including attacks that are only partially resistable
and/or not rendered entirely harmless by reflection (like what the wand balance
proposal suggests), or themed melee attacks (e.g., for red dragons a melee
attack that does fire damage) that cannot be reflected.  Thematic loot would
include gold and gems of course but also some items themed to the element
corresponding with the dragons' color, e.g., fire-themed items (scroll of fire,
wand of fire, ring of fire resistance, potions of oil or booze, and maybe even a
spellbook of fireball) if the dragons are red, poison-themed items if the
dragons are green, acid-themed if yellow, etc.  Dragon halls can start occuring
a few levels before Medusa and should be given priority (for being generated)
over the low-level "easy" special room types.  For example, if a leprechaun hall
would be generated, but the dungeon level is deep enough for a dragon hall, the
dragon hall should be generated instead of the leprechaun hall.

How much of this is already in Fourk?

Statue halls are another possibility: the walls would be lined with statues, a
few of which would be statue traps.  These would mainly be for flavor, however;
the gameplay significance of statue traps that are obvious and can be easily
avoided is basically nil.

Since this special room has little gameplay significance, it should not count as a special room.

Additional types of special rooms should also be added if we can think of more
good ones.  (I like the way Brogue has puzzle rooms, but the specific details of
how they work in Brogue are not a good fit for NetHack, so if we wanted to do
something like that we would have to design them basically from scratch.  As yet
I have not thought through exactly how something like that might work in
NetHack.)

I agree, look to variants for examples of other types of special rooms.

I've never seen a good reason for why there can be only one special room per level. Why not d3-1 or d4-1?

Tribute Level

See above comments, I think adding tributes to other games is a lot of headache for virtually no benefit.

Medusa's Island

Instead of 1-4 levels above the Castle, Medusa's Island now occurs 3-5 levels
above the terminus.  (This means that it can potentially occur immediately below
the Tribute level, if the Tribute level is as deep as it can possibly be and
Medusa's as shallow as it can possibly be.)

See above - I disagree on this, levels after Medusa are not as interesting because the player is usually just trying to get to the terminus. It should be 1-2 or 1-3 levels above.

There are now four distinct versions of this special level.  All four are water
themed, and all four contain Medusa and a statue of Perseus, which contains a
shield of reflection (always), a sack (or in one version an oilskin sack), a
blessed +2 scimitar, and a percentage chance of a pair of levitation boots.

Using the four from 3.6.0? Or others?

Unlike in version 3.4, the random statues on Medusa's level (i.e., all the
statues created when the level is generated except for Perseus) have the usual
chance of containing a spellbook, just like the statues on any other level.  In
conjunction with other changes, this will help to mitigate the strategic impact
of fixing C343-149.

Not sure how well balanced this is. Especially with the Ken Arnold Memorial Library.

Amazing Maze Zone

The levels below Medusa but above the terminus of the Dungeons of Doom will be
maze levels, but they can have corridor widths and/or wall thicknesses other
than 1.  (I have a basically working patch for generating them, although it was
written against an old nicehack and will need to be updated before it will apply
cleanly to anything current.)

See notes above about mazes. The player should not be required to traverse mazes. They are best off as a separate branch or optional branch. Below Medusa should be something that is not maze filler, I would leave it as normal room-and-corridor levels for now.

If any mazes MUST be on the main dungeon, a few levels of them here is certainly better than all of Gehennom, though.

Theseus could be moved to the Mazes of Mystery.

Terminus

The terminus, or final level of the Dungeons of Doom, occurs at the same depth
as in 3.4.  There are several versions.  All of them are considered a
"graveyard" level whether they contain any undead or not, and all of them are
non-teleport and have undiggable walls and floors.

The Castle of Yendor is essentially the same as the Castle in 3.4, except that
the wand of wishing cannot be recharged.  (See Object Behavior, later in this
document.)  This still makes the Castle of Yendor the most wish-intensive
terminus, with a minimum of two wishes (one of which must be wrested) and
possibly as many as four (ditto).  Getting the Castle of Yendor is the new
equivalent of getting the wand at the full (0:3) in version 3.4.3.

This is pretty unbalanced with respect to wishes, compared to the other variants. It's fine to reduce the number of wishes at the terminus, but this contains 2 guaranteed wishes whereas the other variants contain zero. It's all well and good to say this is the equivalent of finding a (0:3) wand at the Castle, but the others might be like finding a (1:2) or (1:1) wand there.

Essentially, the reason wishes are good is because they can be adapted to any game regardless of role, race, or conduct (well, except wishless I suppose), which is something items in the other variants don't guarantee.

Aladdin's Palace: You start on an island near the south edge of the map and must
cross the water (there is a drawbridge to your north) to reach the Palace
proper.  The water contains a dozen random ;.  The central area (behind the
portculis) contains a hostile djinn, four hostile palace guard captains (similar
to regular captains), four hostile palace guards (similar to lieutenants), a
mumak, a monkey, and 18 throne-room monsters, as well as a throne.  Doors at the
back lead to barracks full of soldiers, which may also contain chests (with
normal random-chest contents).  Doors at the east and west ends of the room lead
to the wings.  Each wing consists of a central hall and three rooms (front,
back, and end).  The central hall of each wing contains three random traps, a
trap door leading to the Valley, and two xorns.  One of the two end rooms
contains a chest, sitting on burned Elbereth, which contains gold, about twelve
stacks of gems, two smoky potions, a magic lamp, an amulet of life saving, two
+5 rustproof scimitars, 3d3 random scrolls, two random wands, and three random
items.  The front and back rooms of each wing are store rooms, provisioned with
armor, tools, potions, and spellbooks.  The back rooms each contain an exit to
the outside, and behind the palace there are four dragons and a jabberwock.  The
water also wraps around back, so some of the sea monsters can find their way
there.  A map of this level is included under Proposed Levels, below.

I seem to recall someone complaining about an implementation of this in Fourk because it only has an 80% chance of a wish, which can really leave you hurting if you don't get it.

The Ken Arnold Memorial Library: The player arrives on the level outside the
building and must enter through the double doors at the front (west end).  Most
of the loot is books and scrolls, with maybe a handful of wands.  The chest
contains a pair of lenses, 2-3 markers, and an athame.  There may also be a
scroll of wishing in the chest, if it is decided that it's worth adding that to
the game just to include a wish on this level.  Defending monsters are mostly
spellcasters.  It is dangerous to use conflict here because of the peaceful
Librarians.  A map of this level is included under Proposed Levels, below.

Same thing. Needs that guaranteed wish. Also, it'd be pretty annoying to find this as a Valkyrie or Barbarian who doesn't care about spells. Worse, an illiterate character would get absolutely nothing useful from this.

Spellcasting pack rats might have trouble with this place trying to haul spellbooks back to their stash.

If librarians are definitely being added, the other places in this document that say "if new monsters are going to be added, add this one" should add those monsters too.

Doctor Jeckyl's Laboratory: The laboratory contains a veritable ton of potions
(some random, some of guaranteed type, and some of guaranteed appearance,
including typically 40-50 smoky potions, good for about one wish collectively on
average), a few books and scrolls, a wand each of polymorph and death, and
assorted boots and cloaks.  Flavor items include alchemy smocks, cans of grease,
tinning kits, stethoscopes, a couple of ice boxes, and eleven sinks.  A map of
this level is included under Proposed Levels, below.

Letting the wish method be smoky potions is pretty unreliable. I pity the adventurer who randomly gets smoky potions as sickness or paralysis. Provide a scroll of wishing and don't increase the number of smoky potions beyond normal. Later edit: this actually can give zero to four wishes (more are possible but highly unlikely) with much volatility.

Pack rats would have a lot of trouble with this place from their compulsion to take as many potions as they can carry.

With the below comment about guaranteed wishes, I think a (1:0) wand of wishing and not so many guaranteed smoky potions would be better flavor-wise than a scroll; e.g. Doctor Jeckyl didn't realize one could wrest the wand.

My issue with the variants is that they replace wishes with other items, which may or may not be useful to the adventurer depending on role or conduct (especially true for the Library). And randomly placing a level variant that isn't very useful to the adventurer can be very annoying; the community responses to Orcish Town in 3.6.0 demonstrate that much. There should be at least one guaranteed wish at the terminus, and possibly one non-guaranteed wish. A good method is to provide a scroll of wishing or (1:0) wand of wishing.

Valley of the Dead

The Valley of the Dead now comes in several variants, but in principle they are
all similar.  All versions contain corpses, graveyards, a temple of Moloch,
one-way stairs back up to the terminus, and the bidirectional stairs downward
that are the entrance to Gehennom proper.  The version in 3.4.3 is retained and
is typical.  For branchport purposes, the Valley is considered part of Upper
Gehennom.  The new versions of this level are included under Proposed Levels.

The Valley seems to me to be one of the first places where random generation of substantial parts of the level should be tested. I envision tapering ravines in particular, filled with undead.

Upper Gehennom

The Garden of Temptation is a new level, which has a 50% chance of being
generated in any given game (otherwise, it's replaced with a filler level).  The
Garden features a lake, a number of pools and fountains, numerous trees, statues
(mostly of nymphs and foocubi; there is also one statue of Aphrodite), nymphs,
foocubi, and sundry.  If adding new monster types is permitted, I'd throw in
sirens (dangerous if you're standing close to water, because they can charm the
player and lure him into water, resulting in drowning while helpless) and maybe
also satyrs (basically like nymphs on steroids).  The lake contains a submerged
magic lamp, and some flavor text hints at this.  The nymphs (but only the ones
created when the level is generated) each get a ring, which is either a ring of
adornment (50%) or a random ring.

Seems very rewarding for a level that won't appear 50% of the time. Especially if the statues all have a chance of spellbooks. Imagine if the Big Room guaranteed a magic lamp; that would make 40% of players very lucky and the rest annoyed they didn't get it. I feel like a 100% chance of a magic lamp on a level that occurs 50% of the time would be more frustrating than a 50% chance of a magic lamp on a level that occurs 100% of the time, even though the probabilities are the same.

Either this level needs to be guaranteed in Upper Gehennom, or placed in the Swamp of Death, or have the amount of goodies reduced, or somehow guarantee another magic lamp in Upper Gehennom if it doesn't appear.

What happens if you stone to flesh Aphrodite? What kind of monster is she? Like the Unnethack Aphrodite? Edit: yes, she'll be imported from SLASH'EM or UnNetHack.

Satyrs have a charm attack? Doesn't sit quite right with me.

Vlad's Tower

In addition to increasing Vlad's basic stats (hitpoints, AC, etc.), I recommend
giving him a venom-style "vampire blood" non-reflectable spitting attack that
drains levels (if the target is not drain resistant), which is not blocked by MC
and is usable both at range and also point-blank.  One reason I propose this is
that it has reasonably good flavor justification.  Another reason is, draining
attacks are under-represented in NetHack generally, especially ranged ones, so
having more of them is good.  (However, I don't want to go off the deep end and
adopt Slash'em's wand of draining, because that's just crazy.)  A third reason I
propose this is that it makes Elbereth tactically ineffective against him,
without having him ignore it outright: he'll refrain from melee and instead use
the spitting attack, which drains levels.  With this modification, Vlad is still
pretty much a joke for lawful longsword-wielders, but other characters will need
to take him somewhat more seriously than before.  Beating him to death with a
rusty tin opener will not be so easy.

In connection with the AI change I propose above, Vlad should be able to use such an attack from any square within a 3 square radius. (Animating the attack is then difficult but doesn't really add all that much, I think.) This is to allow him to try to stay out of melee range, but still be able to attack at range.

The first level of Vlad's tower should be bones-eligible (I have no idea why it
wasn't already), and bones should be loaded there unless the player has entirely
disabled the loading of bones.  That is, this level is no longer considered
"special" for bones purposes.

The last level is not bones-eligible because of the Candelabrum.  This could be
resolved, if desired, by removing any extant Candelabrum from the level when
bones are saved (this is already done anyway, in case you happen to have carried
the thing to any bones-eligible level) and then adding it back in when the level
is loaded, at Vlad's location if he is present or a random location on the level
otherwise.  None of the other properties of the level are so important that they
would necessarily need to be preserved in bones.  As usual, bones for these
levels are not loaded by default; the user must set the "bones" option to 2 to
enable loading of bones on traditionally-ineligible levels.

It shouldn't be bones-eligible because making it eligible opens up the possiblity that the player might not face Vlad. Re-generating Vlad if he is not on the level seems silly, and doing that to the Candelabrum too is also silly.

I can't think of a good reason to significantly alter the loot in this branch.
The guaranteed items here are, in any vaguely recent version of NetHack
(certainly since well before version 3.4.0) mostly strategically insignificant
pure-flavor items (e.g., a mundane longsword), which are fine.  The major
exceptions to this are on the middle level, which contains two potentially
important items (life saving, water walking); however, it is my opinion that the
presence of these in Vlad's tower doesn't really alter the overall game balance
much, because there are numerous other ways to obtain them in most games (not
least, they are both commonly achieved results when polypiling).  So my current
thinking is to leave them stand as-is.  However, if more books are needed for
balance reasons, due to the fixing of C343-149, some could be added here.

No more books! Plus, there's already a guaranteed spellbook of invisibility here.

I have elected NOT to add a wish-granting item to Vlad's Tower.  Upper Gehennom,
being within half a dozen levels of the Terminus, really only needs one magic
lamp at most, and the Garden is not only a more flavor-appropriate place for it
but also allows it to be "hidden" better (so that the player must actually do
something to obtain it, beyond just walking in and picking it up) and
furthermore is not guaranteed to be generated in every game, making the lamp
less than guaranteed, which IMO is good.  We should have, in Gehennom, one magic
lamp that is less than guaranteed but significantly more likely than random
generation; and if we are indeed going to have a magic lamp that only appears
half the time, Upper Gehennom is the ideal place for it, since it's so soon
after Terminus, where the user probably just received at least one wish.  This
is potentially a little rough on a player who gets the Library variant of
Terminus (thus no wish there) and no Garden, but I think the better solution to
that (if it is deemed to require a solution) is to manufacture a
flavor-appropriate way to put a wish into the Library; this possibility is
discussed separately.

Also, with the new way we are doing wishes, virtually none of them are actually
guaranteed.  A magic lamp grants only 80% of a wish on average, for instance.
This fact will already need to be taken into consideration when balancing the
new Gehennom, so the possibility of no lamp at Terminus and also no Garden and
thus no lamp in Upper Gehennom doesn't really break anything additional,
provided it's uncommon.  With four variants of Terminus and the Garden generated
half the time, I calculate a one-in-eight chance that the player gets Library
and no Garden; this is very comparable to the probability that the player gets
no wish from Aladdin's Palace or the Laboratory.  So even if we do not put a
wish in the Library, I will still recommend putting the Upper Gehennom lamp into
the Garden and not putting any into Vlad's.

I do agree that slamming a load of guaranteed wishes on the adventurer at the Castle is not the way to go, but having virtually no wishes be guaranteed is not the way to go either. This can make a game very difficult at random.

I think Fourk didn't actually follow this strategy, and it provides scrolls of wishing at various points in Gehennom instead of lamps?

Middle Gehennom

The Fire Pits

Random item generation in the Fire Pits can, if desired, be given different
probabilities from the rest of Gehennom.  For example, it seems logical that
potions would not be generated in the Fire Pits, because liquids (other than
lava) would be in danger of evaporation here.  This would extend to death drops,
as well as floor items and possibly polymorph results as well.  Perhaps even
monster starting inventory should be restricted.

Also paper items? Except for the scroll of fire and spellbook of fireball, of course.

Yeenoghu's Plaza is a roughly square building taking up the middle portion of
the level, stretching across it from north to south.  To the west and to the
east are normal Fire-Pits filler material, containing the up stairs and the down
stairs, respectively.  Secret doors allow access to the interior of the
building, which contains a number of walled rock pillars and a randomized
mixture of lava and floor.  Yeenoghu himself is hiding behind a pillar, in one
of several tactical positions chosen so that the player cannot easily traverse
the building from door to door without being seen at some point.  He is set to
ambush mode, i.e., he waits until he has you in line-of-sight and then attacks
immediately.  (For ambush mode, it does not matter if he can actually see you or
not.  He expects you to come, and as soon as you pass within his line of sight,
he attacks.) The middle portion of the level, including the entire building and
the rock at the map edges to the north and south of it, is undiggable and
unphaseable.  However, if the areas to the east and west of the building happen
to contain rock, it is diggable and phaseable.  Samples of this level are
included below.

Ambush mode is interesting, does this already exist in any capacity?

Dispater's lair is a Gauntlet.  It, too, makes use of the aforementioned "ambush
mode".  The central part of the level is a building containing a winding hallway
with numerous branches.  Each branch contains a demon set to ambush you; one of
these demons (at random) is Dispater.  There are also several "cages" (separated
from you by iron bars), which are infested with things that have ranged attacks
that can go through the bars.  The up stairs are somewhere to the west of the
building, and the down stairs are somewhere to the east, and all the walls and
rock in the building itself (and the rock at the map edges to the north and
south of it) are undiggable and unphaseable.  Samples of this level are included
below.

"Numerous branches" makes me think that there should be different possible ways to reach the stairs, and the level generator will erect random walls in all but one of these, so you have to figure out which is the correct path. Bonus points if the level is unmappable. Edit: The proposed level map doesn't have that, oh well.

Cages make me think of the centaur turrets in one of UnNetHack's Ludios maps.

Swamp of Death

The Swamp of Death is intended to be the more difficult but also more rewarding
path through Middle Gehennom.  In addition to being charactarized by a lot of
open spaces, traversing the swamp means dealing with multiple consecutive
water-intensive levels, with the dangers that entails.

Monster generation per turn will be cranked up significantly in this section;
the flavor explanation for this is, swamps are just teeming with of all kinds of
weird life (and decay, and death, and undeath); the gameplay reason is even
simpler:  this is supposed to be the harder path.

Additionally, the random monsters in this branch lean toward the rather nasty,
and the guaranteed demon-lair bosses are also more dangerous than their
fire-pit-dwelling brethren.  Monsters you can expect to see a lot of in the
Swamp include sea creatures (; and N plus water demons and water trolls),
couatles, dragons, water and air elementals, decay-themed monsters (gas spores,
green slimes, jellies, and molds), snakes (especially pythons and water
moccasins), and certain demons (notably nalfeshnees, balrogs, foocubi).  There
may also be random g, both kinds of mind flayers, random L, random &,
crocodiles, chameleons, sandestins, purple worms, random i, vampire lords,
random O (living in swamps as a subtle reference to Shrek), water nymphs, and
occasional titans.  Salamanders may also be included in the Swamp as a bad joke
(because of the real-world amphibious creature by the same name).

Air elementals seem a bit misplaced to generate specifically here. Sure they beef up the monster difficulty, but I don't think of swamps as windy or airy places.

As in vanilla Juiblex's Swamp, there should be all P and b too. Also perhaps keep lichens since they are a decay-themed monster.

The Swamp of Death contains two new types of traps that do not occur elsewhere.
The first and simpler of these is the anoxic pit, a type of pit whose interior
is lacking in oxygen, due to the accumulation of swamp gas.  Falling into an
anoxic pit starts a short timer, which will kill you by asphyxiation if you
don't get out of the pit.  (It is not necessary to leave the tile; you just have
to get out of the pit.)  Being unbreathing or having magical breathing makes you
immune to this effect.

Why not just have all normal pits be "anoxic" if generated in the Swamp? I don't really see the need for a new trap.

The other new type of pit in the Swamp of Death is quicksand.  When not known,
quicksand looks like normal ground, but if stepped on it behaves in much the
same way as lava (sans fire damage), drowning you on a short timer if you don't
get out.  Like lava, levitating works great if you aren't stuck but will not
unstick you if you are.  Unlike lava, quicksand cannot be converted into normal
ground by freezing it.  However, zapping a wand of digging downward will convert
it into a hole, dropping you and any sunken items through to the level below.
Teleport (if the level allows) or phasing (ditto) will get you out.  Unbreathing
or magical breathing will prevent you from drowning in the quicksand (buying
time so that you can e.g. dig down with a pick-axe, which would otherwise be too
slow) but will not unstick you directly.  Quicksand cannot occur on levels with
an undiggable floor and is only generated in the Swamp of Death branch.

Levitation, if it won't unstick you, should stop the timer. This goes for lava too.

Can items be generated on this?

I kind of feel like both the Fire Pits and the Swamp encourage the use of levitation too heavily, though.

It is tempting to add a type of monster (other than the unique named demons in
their layers), generated only in this branch, with an illness attack; but I'm
not sure whether that is really necessary.

There aren't enough illness attacks in the game. Go for it.

Lower Gehennom

Continuing downward from either the Fire Pits or the Swamp of Death will land
the intrepid adventurer in the Black Market, which is to be imported from
Slash'em and/or UnNetHack and modified to suit.  One important modification is
that there will be two up stairs, one leading to each of the Middle Gehennom
sub-branches.  Another modification is that the level must be traversed (if
nothing else, the Amulet must be carried across it on the ascension run).  Both
of the up stairs will be at the western end of the level, and the down stair
will be near the eastern end.  Prices in the Black Market are obscene enough to
allow the player to spend all the gold in the game on just a small percentage of
the total inventory; but Gehennom contains enough gold to allow the player to
buy items of particular interest (e.g., missing pieces of the ascension kit).
The Black Market is specifically guaranteed to contain at least seven wax
candles, just in case.

Some additional changes to the Black Market will be needed to make waltzing out
with with all the stuff scot free considerably more difficult than in Slash'em.
Games in which the player does so are intended to be very much the exception
rather than the rule.  If we balance the Black Market as intended, the player
will in most games be able to buy a couple of things but leave the rest of the
inventory on the floor and the shopkeeper peaceful.  To this end, the entire
level is considered part of the shop.  The level is non-teleport for the player.
There is more than one keeper so that they can cover at least both of the up
stairs ("zone defense", as it were).  (Two keepers could be a married couple, or
three could be siblings, allowing the down stairs to be covered as well.)
Additionally, the keepers can teleport at will (even though the player cannot),
and if there are unpaid bills they go straight for the stairs.  Cursed gain
level won't work in the Black Market, because it's the top of the Lower Gehennom
branch.  Zapping a wand of digging downward will make the staff hostile, and you
have to come back up through eventually in order to ascend; the keeper(s) will
certainly remember you.  The keepers will have a metric trainload of hitpoints
and a full set of intrinsic resistances (including slow digestion, to thwart
purple-worm antics) and crazy-high monster MR (high enough to always resist
conflict, as well as polymorph).  Black-market shopkeepers are incredibly
dangerous when hostile, capable of one-shotting many late-game characters.
(Perhaps at some point in the backstory these shopkeeper each ate fifty rings of
increase damage then killed Chuck Norris and Kirby and absorbed their powers, or
something.) Also, Black Market shopkeepers are immune to stoning and genocide.
Basically, there is not supposed to be an easy way to kill them.  Pet-based shop
schenanighans should also be thwarted if possible, especially ones that
otherwise would trivially allow the entire inventory to be cleared out.
(Perhaps the stairs should be blocked if any monster on the level is carrying
unpaid black market inventory, but in that case it would be necessary to think
through the balance implications of hostile covetous monsters that might pick
things up.)

I think the Black Market isn't really very necessary; there are a ton of special cases it creates, and the only important gameplay benefit is you have to lug a lot of gold there to maybe get a needed item in your ascension kit. Seems contrary to the sentiment that you shouldn't need much gold after you finish buying protection.

Also, it doesn't work for flavor reasons. Why is the most powerful monster in the game the shopkeeper in the black market? Even in UnNetHack, One-eyed Sam is less scary than Cthulhu and Demogorgon (though still scarier than the Wizard, which I find problematic). Why is the black market so deep in Gehennom? Who else shops there? If it's at such an important junction, why don't the Gehennom residents kit themselves out there? How do they manage to make sales often enough to stay in business if an item costs a significant fraction of all the gold in the entire dungeon? Why couldn't they just ascend on their own if they're this powerful?

There should be some special level at the top of Lower Gehennom, but not the black market. If it must be in the game, it shouldn't be required to traverse.

Moving on from the Black Market, the down stairs lead into the rest of Lower
Gehennom.  The new cavern generator that ais created, which I have lightly
modified (mainly to add regions of water and lava) is designed so that it can
generate around an embedded special center.  In Lower Gehennom we take advantage
of this to embed all three levels of the Wizard's Tower as well as the Fake
Towers.  There will also be several pure filler levels in this section, which
will be handled entirely by the new cavern generator.  Because of the greater
dungeon depth here, the probability on any given level of water and especially
lava is much higher than in Upper Gehennom.

The Wizard's Tower has been partially redesigned.  Several features of the
original Wizard's Tower, notably including the zoo on the middle level, were
clearly intended to be obstacles that must be overcome as victory conditions but
in practice were much too easy for any character sufficiently advanced to be
even contemplating the possibility of awakening the Wizard.  This problem may
not be entirely solvable, but that has not stopped me from trying to at least
mitigate it somewhat.  Additionally, the physical size of the tower itself has
been somewhat reduced, to allow greater variety in the surrounding filler.

The zoo could be filled with nasties. Edit: The plan is for the zoo to be filled with demons. Fine. And traps. Not fine, see below.

The Wizard of Yendor, when he first appears at the top of his tower, is carrying
a magic lamp and a magic marker.  Each time he appears again (after you've
killed him; but NOT counting duplicate appearances from Double Trouble) he
carries at least one candle and at least one spellbook.  The book is mainly for
flavor.  (The player probably has all the books they need by now, but it just
"feels right" that the Wizard would be carrying such a thing.)  The candle
however serves a purpose: this is another way the player can always obtain
enough of them to do the invocation, in case killing dwarves in the Mines was
undesirable, or perhaps some players will want to brag that they got all seven
of their invocation candles by killing the Wizard repeatedly.

How many candles does he respawn with? 1d3?

Not in his first incarnation, but in subsequent appearances, he should have a chance to be wearing one of these items: an amulet of reflection, a shield of reflection, or a cloak of magic resistance. If not wearing the amulet of reflection, he then has a second, small, independent chance to get an amulet of life saving. He's more of a threat if you can't instakill him with a death ray.

There is also a guaranteed magic lamp in Lower Gehennom itself, in the Fake
Tower that doesn't have a portal.  It is guarded by a new always-hostile unique
named demon created especially for the purpose.  Possible names for this demon
include Therion, from a Greek word meaning beast, Daimonion, from the Greek word
for demon, Mashchith, from a Semetic word meaning destroyer (though that one
might ought to be an A rather than &), Gozer, from a popular eighties movie, or
we could take the opportunity to reference another roguelike game or a worthy
work of fantasy fiction that NetHack doesn't already reference.  This special
guardian, generated asleep or meditating, should be rather fast and have a fair
quantity of hit points and be capable of spellcasting and also have powerful
melee attacks (enough to be dangerous to even a well-buffed late-game player
character, if not treated with proper respect).  It should be resistant to
stoning, teleportation, polymorph, and at least partially to wand/finger of
death, should have intrinsic slow digestion (so a purple worm won't eat it), and
if possible should be carrying the magic lamp in its inventory before the player
arrives.  Digging into the tower should wake it up, if that isn't too much
trouble to implement.

Generate this guardian next to the player when the player picks up the lamp or enters the tower. The player shouldn't be able to use telepathy to identify the tower as the fake one. If the player shouldn't be able to use object detection either, generate the guardian with the lamp in its inventory when the player steps into the center of the tower, unsure of whether the portal is there or not.

Intrinsic slow digestion is overkill; if the player manages to get and level up a purple worm capable of eating the guardian, let it.

Moloch's Sanctum

In the first place, most of the riff-raff in the Sanctum is to be cleared out.
There will no longer be a dozen unaligned aligned priests casting Summon Insects
constantly.

Yes, although I think the priests are fine as long as they don't cast summon insects. Paralyze, tower of flame, open wounds, geyser, and curse items are all quite good spells and present a decent challenge we shouldn't do away with just to get rid of the ants.

In their place, we have the High Priest of Moloch, who will be buffed until he
is an actual threat.  In addition to better resistances, more speed, and other
basic stat buffing, he is to receive a special attack, unique to him, that
causes random tiles in the general vicinity of the player to become lava.
Preference is given to floor tiles with no trap on which nothing currently is
standing, but if those run thin on the ground, other tiles are fair game as
well.  (Stairs, ladders, portals, high altars, and rock and walls that are both
undiggable and unphaseable are exempt.) For obvious reasons, the High Priest of
Moloch has permanent intrinsic lava-walking (as from fireproof water-walking
boots, but inherent to his person).

Resistances include magic resistance and reflection. Also shock resistance, so when Moloch blasts the player with lightning for harming his servant he doesn't kill said servant (this is a protection against future effects of divine lightning that might bypass reflection).

Lava attack means that the player will probably just quaff a blessed potion of levitation or cast skilled levitation beforehand. The attack is not a bad idea, but if he spams it a lot, every player will feel like they should do this. I think the trick is to make him not use it enough that a lot of the surrounding terrain will become lava in a fair length fight.

Nah, give him fireproof water walking boots instead of a new resistance. If the player actually manages to get them stolen off him, let him fall into the lava. More interesting that way.

Additionally, the High Priest now has a melee attack that does shock damage.
This can be partially resisted, but it can still potentially destroy rings and
wands.  He can also cast the Destroy Armor monster spell, destroying a random
piece of the player's worn armor, as well as Curse Items.  Whether he should be
able to Summon Nasties and/or gate in demons is an open question.  Finally, the
High Priest also now has a confusion attack.

Does this destroy armor spell bypass magic resistance? Because otherwise it won't do anything.

I'd change it to stunning rather than confusion, it's a worse effect that still impairs spells, etc.

The physical layout of the level itself is of relatively minor importance, but
out of thoroughness I have included two suggestions for new variations on the
layout.  All three versions of the level (the one from 3.4.3 and the two new
ones) are divided into two main areas: a series of connected smaller rooms and
corridors, on the east side, and a much larger, roughly circular room on the
west side with a rectangular temple embedded.  The up stairs are always in an
"entry area" near the southeast corner of the level, and a secret door leads out
to the hallways and small rooms.  In all three cases it is possible to skip some
of the rooms, but it is not possible to skip the graveyard room, which is always
the largest of the rooms on the east side of the map.  A vertical stripe of
unphaseable rock separates the east and west sides of the map, penetrated only
by a single corridor, which leads to the larger room on the west side.  The
temple itself is always surrounded by fire traps and contains the high altar and
the high priest, who waits for you to enter the temple and then becomes hostile.

I thought the graveyard in the Sanctum was being removed as part of the "riff-raff": it's mostly stuff that goes down in one or two hits (zombies, mummies, wraiths) and if it gets awakened, all it really does is block the player from returning to the stairs.

The Elemental Planes

Genocide is now ignored on the Planes.  The flavor reason for this is simple:
you have left the mortal world and entered the elemental planes, the very fabric
from which the universe is constructed.  Monsters that are genocided are wiped
out from the entire dungeon, perhaps even from the entire world or universe, but
the Planes are not part of the dungeon or part of the world and, indeed, not
part of the physical universe.  You have left the Mortal Plane.  The gameplay
reason is, otherwise, everyone would just genocide elementals before doing the
ascension run.  Also, genocide is very powerful, and this is a way to gently
nerf it just a little.  Via genocide you can still play the rest of the game
without facing your personal bugaboo, whatever that may be, but if you want to
actually win the game, you have to face your terrors on the Planes.  This is,
after all, the end game.

Fair enough. So this would be implemented not simply as scrolls of genocide having no effect on the Planes, but rather when generating a monster on the planes, don't respect its genocided status. (What about extinction though? Still respects that? It probably should respect extinction, but I could go either way on it.)

The Plane of Air requires few other changes.  I believe it is working largely as
intended.  For consistency with the other planes, earth-based actions (such as
reading a scroll of earth) do not work here (and any boulder that arrives via
the portal instantly disintegrates).  Similarly, being on the plane of air is
instantly fatal for earth elementals (they lose their cohesion and disintegrate,
or something; in the case of polyself, you revert to your natural form).
Additionally, it may be that some of the monsters on the plane may have
lightning attacks that can destroy wands and rings, even if you have reflection
and shock resistance.  (Balance testing will be required to determine whether
this change is viable.)  Air elementals here are stronger than anywhere else in
the dungeon; in practice this mostly means they take more hits to kill.

I would implement moving clouds that obscure vision, but do not display the # character on blocked-sight tiles behind the cloud which are not part of the cloud itself (which I find really annoying). This is, I think, what 3.6.0 has with just that small rendering fix. Static, nonrandom clouds are boring.

The Plane of Water needs several changes.  In the first place, it is no longer
possible to levitate "over" the water outside the bubbles, since the entire
plane is filled with the water.  When in the bubbles, you need levitation or
water walking or flying or floating or swimming (or a steed that has one of
these properties) or magical breathing to avoid drowning, since you are on the
interior surface of an air bubble, with water beneath you (or, in the case of
magical breathing, you would be submersed).  Thus, the bubbles on the plane of
water are like water terrain elsewhere in the game, and the area outside the
bubble (which ideally should have a distinct glyph, or at least a different
color) is surfaceless water with no air above it, so anything occupying those
tiles is fully submerged at all times.  With this change, swimming monsters (or
ones that are capable of swimming to the surface, at any rate) can swim right
into the bubbles unimpeded, and flying monsters (e.g., couatles) can only occur
within the bubbles but cannot leave them.  Secondly, the portal no longer moves.
Its position is randomized, but it is stationary.  As mentioned above, nothing
that relies on fire will work on this plane.  This includes most light sources
(Sunsword works fine, as do spells and wands of light; magic lamps are
negotiable, depending on whether we consider their ability to give light to be
part of their magic).  It also includes wands, scrolls, and spells of fire; fire
horns; and all attacks that do fire damage, whether from monsters or the player.
Fire traps are not generated, and the fire-trap effect of magic traps does not
occur.  Fire elementals and fire vortices and flaming spheres (and possibly also
salamanders) cannot survive on the level:  entering the plane of water is fatal
for them.  Their inner fire, which is their life force, is extinguished.  On the
other hand, water elementals, not really a threat elsewhere, become on the plane
of water a force to be reckoned with.  Normally slow, they get 36 movement
points per turn when in the water terrain that only occurs here (outside the
bubbles only; inside the bubbles, they get their normal amount of movement
points).  Additionally, their melee attack, when used anywhere on the plane of
water, does double damage and can also deal a staggering blow, stunning and
knocking the target back.  Beware of being knocked out of the bubbles (if you
don't have a way to survive that) and be careful about getting surrounded.  I
was tempted to also give them an engulfing attack (usable only on the Plane of
Water) that drowns you on a short timer (unless you are unbreathing or magical
breathing), but that can be held in reserve pending play testing -- if the level
is too easy without it, it can be added, otherwise not.

There needs to be some way to safeguard the player entering the plane who does not have levitation, flying, water walking, or swimming from drowning immediately. That has the potential to be one of the worst YAADs, particularly because you don't know whether the Plane of Water is the one coming up next. It'd also be basically impossible to learn without either reading spoilers or getting a splat. I feel that all the portals should drop you in "safe" areas where you can collect yourself before proceeding.

If the portal is in a fixed location, it should be surrounded by a nonmoving bubble. Bouncing around in the bubbles is annoying enough without the portal being located in inaccessible water that the bubbles keep dodging.

I think the player should be able to direct the bubble's motion somehow, but I can't think of an appropriate mechanism.

The plane of fire is completely different from the one in version 3.4.3.  That
version wasn't a bad design, but now between the Fire Pits and the High Priest
of Moloch, lava has become old hat for a player who reaches the Planes, so the
plane of fire now instead features a new kind of terrain, more in keeping with
the other planes: this terrain is creatively entitled "fire", and being in this
fire terrain is indeed like being on fire (not just on the surface but to the
core of your being) and entirely surrounded by flames.  Being in this fire
terrain does fire damage every turn, which is only partially resistable for the
player.  Non-fireproofed items (including ones in your inventory) are damaged
just by being here.  Potions boil.  Paper objects burn.  Taking a potion of full
healing out of your bag, with the intention of quaffing it, will simply result
in a boiled potion.  Similarly, taking out a scroll of portal detection isn't
going to help you.  Keeping items in a container does protect them.  Fire traps
are not needed here, because they would be redundant.  Being suspended entirely
in burning flame also means there is no floor, much like on the plane of air,
with similar implications for movement.  Additionally, there are now clouds of
smoke on the plane, similar in principle to stinking cloud except that they do
not time out, don't do damage directly, are opaque (i.e., they block vision),
and are non-stationary, drifting in a manner similar to the bubbles on the plane
of water.  The fire elementals, for their part, get 36 movement points per turn
when in the fire terrain (whether or not the tile has a smoke cloud; i.e.,
pretty much everywhere on the plane) and are also harder to kill here than
anywhere else in the game.  Their attacks are only partly resistable, but the
real danger is that they will block your path and prevent you from reaching the
portal before the terrain itself burns you to death.  Fire elementals, because
they do not show up on telepathy, can easily sneak up on you here unless you
have warning (or stay clear of the smoke, which may not be practical).  No
monsters can be generated on this plane unless they are fire resistant, and even
fire-resistant monsters take damage from the terrain unless they are basically
made out of fire or otherwise strongly themed with it (flaming spheres, fire
vortices, fire elementals, probably  salamanders, maybe red dragons).  Fire
vortexes can be generated, even though the plane is clearly outside of Gehennom.
Entering this plane is instantly fatal for water elementals (they boil away to
nothing or something), and the fire damage will soon kill anything that's not
fire resistant, how long that takes depending on hit points.  This does create
difficulties for getting most pets through the plane alive.  I'm not sure if
that counts as a balance problem or not.  It doesn't make getting a pet through
impossible, just challenging.

I would make the fire terrain not universal. Use the island generation algorithm to generate "islands" of either solid floor or empty air, in which you can take out flammables, park pets, etc. The entry portal deposits you on one island and the destination portal is also on an island.

Smoke is an improvement over 3.6.0's annoying poison clouds, which are only there because of the lava anyway.

The plane of earth is conceptually similar to the one in version 3.4.3, but
there are a couple of new challenges.  First, any non-rock terrain on this plane
is subject to cave-ins, which can happen at any time, converting it back into
rock.  If this happens on the tile where the player is standing, the player can
become embedded in the rock, in addition to sustaining damage.  Wands of digging
are not considered to be air-based (they're magic based), so they do work here,
and zapping one at yourself will free you from being embedded.  Alternately, if
you dig an adjacent tile, freeing yourself by walking out works the same as on
any other level.  Earth elementals, unlike their counterparts on the other
planes, remain slow here; but they get a large boost in the hitpoint department
when generated here, and their melee attack, when used on the plane of earth,
causes stoning (the delayed version, like from a hissing attack; this effect of
their attack does not occur anywhere else in the game, only on the plane of
earth), so the player will want to carry lizard corpses or have some other way
to deal with stoning.

Cave-ins happen on non-rock AND non-floor (and non-portal) tiles. The generated bits of floor should be safe, else they'd all collapse before the player ever got to them.

Cave-ins are also quite annoying in Grunt, I'm not sure how much I like them.

The last of the five planes is always the astral plane, and it does receive a
few enhancements.  The "Summon Insects" motif is abolished.  Nothing could be
more anticlimactic than to finish out the very end of the game surrounded by
ants.  The aligned priests can stay for flavor reasons, if they can be kept from
filling the level with insects; and the random A are still present; but the real
challenge on the Astral Plane comes from other sources...

Easiest solution is to block the AI from summoning insects on Moloch's Sanctum and the Astral Plane; I think it works well enough on other levels, if you anger an aligned priest or in the Priest quest, for instance.

The "player-character" monsters on the Astral Plane are now always of quite high
level, generated with stats and equipment comparable to what you'd expect a
typical player to have at this stage of the game.  Most have reflection and a
full set of elemental resistances; many also have player MR (either from GDSM or
cloak), low AC, high HP, fully enchanted weapons, and so forth.  Any objects
that the AI knows how to use, which a late game player would likely possess,
should be given to these characters: amulets of life saving, potions of full
healing, chickatrice corpses, and anything else the AI can use against the
player can all be in the offing.  If the wand balance proposal is implemented
(so that reflection and resistances don't make attack wands completely
non-threatening to the player), then these players should each get several
attack wands.  If a ranger character is generated, said character should
multifire from a large quiver of +7 arrows.

Lots of this already exists in vanilla, right? The highly enchanted armor, weapons, even some artifacts. Chickatrice corpses may be a bit too evil, if they can still instapetrify the player. I would look at variants like FIQHack for ideas on generating threatening player monsters.

But the big threat will be the other three Riders.  In the first place, they now
have actual hit points.  A normal player should not be able to kill them in two
hits, even if wielding the blessed +7 Excalibur and a blessed +7 katana with
expert skill.  The riders' speed is also increased, perhaps to 24.  (If it
wouldn't be a major hassle to implement, I'd say they should be riding unique
colored warhorses; but at the very least they should have speed equivalent to
someone riding a warhorse.  Otherwise, why are they called riders?)

Are non-Death riders still vulnerable to wands of death? If so, they won't be much of a big threat. They should probably resist death magic, have player-style MR (I wouldn't suggest that since it makes magic missile ineffective), or have reflection.

Speaking of which, there is no separate rider named Pestilence.  (There isn't
one in the source material.  Pestilence is simply one of the tools the Death
rider uses.)  Instead, the first rider will now be implemented, Conquest
(a.k.a., the rider on the white horse).  Conquest carries a bow and has a unique
multifire arrow attack, usable both at range (unlimited range, as long as
there's a straight-line path) and also point blank, that never runs out of ammo
and always treats the target as having an AC of 0, regardless of the target's
actual AC.  In order to simplify the implementation of preventing anyone else
from using his attack, these special arrows always break on impact, whether they
hit the player or a monster or a wall or the ground or whatever.  (From an
implementation perspective, this is similar to venom, which is never left laying
around as an object on the floor that the player can mess with.)  The number of
arrows Conquest fires per action, and the amount of damage they do, is subject
to balance testing; but it should be enough to make him just a little scarrier
than the new improved Death.  Note that Death is now the only rider that must be
adjacent in order to attack.  The standard strategy will likely involve never
being adjacent to Death, never being in a straight line with Conquest, and being
satiated before approaching Famine.  Of course, this may be more easily said
than done.

Not sure I like Conquest as described here, though making one of the riders a straight up normal-weapon juggernaut is an interesting idea. I think the main reason is because War and Conquest are very similar, whereas Pestilence's sickness attack is more unique and scary, as it's the only one with a delayed instadeath. Plus, despite the source material, popular culture has had the four riders named as Death, Famine, Pestilence and War for a while.

There are also now three major ways to successfully complete the game on the
Astral Plane.  The traditional way is to offer the Amulet of Yendor to a
(preferably co-aligned) diety in exchange for immortality.  A second and more
difficult ending involves keeping the amulet for yourself, establishing yourself
as master over all other mortal creatures (and, particularly, the ones on the
Astral Plane).  The third possibility is to destroy the amulet, which can only
be done on the Astral Plane; this breaks the power of the gods over mortal
creatures forever, so in some ways it is the most satisfactory ending, although
it does not provide your own character with immortality.
The traditional win, by offering the amulet to a diety in exchange for
immortality, is unchanged from version 3.4.3.
To keep the amulet for yourself, it is necessary to assert your dominance over
the Astral Plane by killing everything.  The other Riders do not have to be dead
all at the same time, but you have to kill them each at least once, and there
cannot be anything else left alive on the Plane.  When these conditions are met,
#invoke the amulet to claim it.  You don't get an invisible choir or bathed in
radiance, but the other three Riders cower in terror before you.  Your diety,
rather than granting you immortality, curses your name and attempts to strike
you with lightning and a wide-angle disintegration beam, but if you survive
these trivial attacks you become the dominant power in the universe, for as long
as you can hold onto it.

"I have come. But I do not choose now to do what I came to do. I will not do this deed. The Amulet is mine!"

To destroy the amulet and set all mortals free from the gods' divine
interference, simply sacrifice each of the three high priests on his own altar
(surviving any incidental consequences of doing so), then gather the other three
riders around you (they must each be within your line of sight and alive; but
it's alright if they have been killed previously, as long as they aren't still
dead) and #invoke the amulet.  Dieties of all three alignments will smite you,
but if you survive that, the other three riders will join you in defying the
gods, and the amulet will be broken, taking much of the gods' power with it.

"simply", pffft

Later note: there should be some rumors and oracles added to hint at these endings so the player can find out that they actually exist.

I like these additional win conditions. Particularly since they don't make the standard ascension any harder. The other two remind me a bit of the dNetHack anachrononaut, who can choose to save the future by destroying one of the other high altars.

Object Probabilities

Additionally, the following already-rare items should be made somewhat more
common in Gehennom:
 [...]
 * crystal balls, if they can be made useful for something other than
   polyfodder; I believe somebody had a proposal for that, but I don't recall
   the details.

With the Plane of Fire as proposed here, a crystal ball is virtually required for those who can't cast detect unseen.

Object Behavior

Wands of wishing are not rechargeable.  They are generated with the same 1-3
wishes as before, so with the wrest you get 2-4 wishes from a wand of wishing,
and that's it.  The reasoning behind this is that after getting 5-7 wishes, you
basically want for nothing and the rest of the game can't really offer you
anything that you want or need, but you have to slog through the rest of it
anyway.  Boring.  Two wishes will generally leave you wanting more, and even
four may leave you hankering for another marker or something, so there's still
something you can hope to find during the remainder of the game.

The clean way to do this is make any recharging of a wand of wishing blow it up, and generate wishing with a charge count of 0 (because it indeed has not been recharged). However, I can imagine arguments for consistency with vanilla and variants, making it explode when the charge counter increases beyond 1.

New Objects

I also propose that a handful of new random-appearance items be added that can
ONLY be found in Gehennom (or via wishing or bones), to ensure that the player
will still have meaningful identification to do.  Some of these items should be
useful, and some should be potentially dangerous.

Careful attention will have to be paid here to balance, and so I am refraining
from making specific suggestions on the theory that item details are really a
separate issue from the dungeon overhaul.

Adding more stuff to identify in Gehennom isn't that much of a challenge when most spellcasters will have got their hands on a book of identify by now, and even those who don't will probably have or be able to make a few blessed ID scrolls.

Non-Teleport Levels

Overall, fewer non-teleport levels will need to be traversed when walking from
the top of the dungeon to the bottom, or from the bottom to the top.  (This will
greatly reduce needless tedium for players who make multiple trips back and
forth, e.g., ferrying items to and from a stash.)  As a consequence of this, the
ascension run will need to be rebalanced; but that probably needs done anyway;
frankly the final run was already not terribly well balanced in 3.4.3; this will
be exacerbated if the mysterious force is defanged, as several people are likely
to propose, and so rebalancing will be required.  (One obvious possible
rebalancing measure would be to have the Amulet suppress teleportation within a
level, in addition to levelporting; however, I am not convinced that is the best
approach; it should hopefully be possible to do significantly better.)

The mysterious force will have to be modified as per this version anyway; what if it hits you in Orcus Town? Do you go to the Fire Pits or the Swamp of Death?

From a frustration prevention standpoint, replacing the mysterious force with the Amulet blocking intra-level teleport is a big win. The force feels like a kludgy way of drawing out the ascension run, and I think it'd be accurate to say most everyone hates getting booted back down to lower levels. Meanwhile, players like me stockpile teleport scrolls and leave them at the Castle, so once we reach that stash we can be all the way up through the Dungeons of Doom in fifty turns, not counting crossing Medusa. Making this replacement makes traversing Gehennom on the ascension run take about as long as the Dungeons of Doom, which is a good thing. If it's still too fast, something like Unnethack's making monsters spawn on the upstairs specifically is also a possibility.

The Valley of the Dead is an open question, hinging largely on whether we
consider it acceptable to encourage players to eliminate the unaligned priest
there, effectively penalizing those who do not.  My instinct is against this,
because I feel that it is bad game design to penalize the player for not killing
a peaceful monster.  (It's bad enough that killing a peaceful priest is the only
reliable way to get a robe without spending a wish, not to mention the books
that they tend to carry.)  However, I can see the argument both ways,
particularly since the Valley is always directly below the Terminus, which would
make two permanent non-teleport levels in a row, which is annoying, particularly
being positioned right smack dab in the middle of the dungeon where they are
typically traversed many times per game.  I'm torn on this one.

The non-teleport restriction could just go away once the player gets "So be it." and enters Gehennom proper. This is an easier "completion" condition than the Terminus.

It's also annoying how you can't levelport down past it already, so e.g. going from a stash high in the Dungeons of Doom to a low level in Gehennom requires two levelports. Maybe this levelport restriction could be removed once you pass into Gehennom as well.

It would also be interesting if passing into Gehennom was the trigger to remove non-teleport status at the Terminus, though it's probably not a good idea.

Sokoban is an interesting case, because on the one hand it is a branch (and
theoretically an optional one at that), and so it would never need to be
traversed, especially once it is completed.  In some ways, it is perhaps the
single most obvious case where the non-teleport restriction is crucial to a game
mechanic, and so it makes sense that the entire branch is categorically
non-teleport.  On the other hand, a lot of players like to put a stash there,
and so they do end up traversing it (or portions of it) repeatedly.  On the
gripping hand, some players like to take advantage of its non-teleport status
for dealing with certain kinds of monsters (e.g., unicorns).  Also, the levels
are small enough that traversing them on foot (once the boulders have been dealt
with) is easy.  Additionally, there's no boss monster, so the "take away the
restriction when the boss is killed" mechanic does not fit here.  An additional
mechanic would be needed, unique to Sokoban, for removing each level's
non-teleport restriction when the puzzle is completed.  There are arguments to
be made either way, but on the whole I think my preference is for the whole of
Sokoban to remain permanently non-teleport.

Given that in 3.6.0 the luck penalties are removed when the puzzle is completed, this trigger already exists, and it's pretty popular. I'd remove the restriction once the puzzle's finished. To reply to the argument that it's a strategically significant non-teleport level, Sokoban isn't really intended to be a place where unicorns or other teleporting monsters can be easily hunted; that's just a consequence ofthe fact that it's been non-teleport since its creation.

Wish Sources

The goal is to provide the player with roughly the same number of wishes as in
3.4 but to spread them out more.  Currently I have the following:

Terminus (1.25ish): Castle of Yendor contains a non-rechargeable wand.
Aladdin's Palace contains a magic lamp Dr. Jeckyl's Laboratory averages 40-50
smoky potions and a magic marker.  Ken Arnold Memorial Library might not have
any wishes, unless we want to implement a scroll of wishing; but it contains at
least two magic markers

Upper Gehennom (0.5): The Garden of Temptation (present in 50% of games)
contains a (submerged) magic lamp.

Middle Gehennom (1): Demogorgon's Bayou Village contains a magic lamp

Lower Gehennom (1+): The Black Market is likely about as good as a wish.  One
Fake Tower has a magic lamp (the other has a portal).

Wizard's Tower (1): The Wizard of Yendor, the first time he appears, carries a
magic lamp and two markers.

It is tempting to also add a second wand in the very late game (perhaps in the
Sanctum, or have Rodney drop it the second or third or fourth time he is
killed).  The reasoning here would be to provide "emergency wishes" for use on
the five Planes, when spending multiple turns rubbing might be out of the
question.  However, I am not sure whether this is necessary or even a good idea.

My analysis of the wishes as this proposal outlines them (the above does not account for the 20% chance of a magic lamp not granting a wish):

• Terminus averages: 3 / 0.8 / 0 / (difficult to calculate and benefits vary based on random smoky potion's effect, see below). Guaranteed wishes: only 2 if the Castle of Yendor is chosen.
• Upper Gehennom: 0.4, no guaranteed
• Middle Gehennom: 0.8, requires passing the Swamp and fighting Demogorgon to get it, no guaranteed
• Lower Gehennom: 0.8 plus black market, no guaranteed
• Wizard's Tower: 0.8, no guaranteed

Advanced wish calculations:

Demon zoo in the Wizard's Tower, assuming it doesn't specifically block wishing from being generated: It could make a few players wildly lucky and generate 1-3 wands of wishing, but the average player wouldn't find one. Among 63 randomly generated wands, it's a 72.9% chance of none being wishing, a 23.1% chance of one, a 3.6% chance of two, a 0.37% chance of three, and a 0.027% chance of four. Yes, it's only a little less likely for the player to find three random wands of of wishing in the zoo than it is for a single random wand on the floor to be wishing. Probably the zoo is intended specifically not to generate wands of wishing and that just isn't described here.

Dr. Jeckyl's Laboratory: Assuming the player collects 50 smoky potions and no djinni have been created yet, and also accounting for the 20% chance that a djinni from a blessed smoky potion does not grant a wish, the probability of getting no wishes from quaffing all 50 potions is 4.17%. If the player gets 40 potions, the chance of no wishes is 7.9%. The probability of getting one wish from 50 smoky potions is 13.7%; from 40 potions it is 20.7%. Since the chance of getting a djinni decreases with each one created, a simple binomial distribution isn't good enough to calculate further probabilities, but suffice it to say that the Laboratory has the most volatile number of wishes, with two, three, four and even five-wish outcomes remaining statistically significant. This is exacerbated by the fact that the wish delivery method is potions, which have their own randomly assigned effects on the player and on monster behavior.

There need to be some guaranteed wishes in here somewhere.

Closing Remarks

I have not, in this proposal, attempted to reddress every problem with the game.
Several things that are quite thoroughly broken are not addressed here at all.
The player's alignment record, for example, was clearly supposed to be a
significant factor in the game, but in practice its only real impact is to place
a lower bound on speed runs.  The need to eat is intended to force the player
deeper into the dungeon as food supplies run low, but in practice it only ever
does this in the very earliest part of the game, because later on food is
essentially infinite.  (If you want to see a roguelike that gets this *right*,
play a few games of Brogue.  The specific way in which it is handled in Brogue
is not applicable to NetHack, because the two games are different in a number of
other ways; but the fact that another game does get this right demonstrates that
it is possible.)  The manner in which score is kept is arbitrary to an extent
that makes score almost entirely meaningless.

The food problem could be addressed by only generating permafood items on levels above the Oracle or the Quest or something, and when polypiled and generated from horns of plenty and so on.

Perhaps the worst of all is the difficulty curve, which is entirely the wrong
shape, almost exactly the opposite of what it should be.  The problem with game
difficulty is not that it is too hard, nor that it is too easy, but rather that
the most difficult part of the game is the first part, and the easiest part is
very late.  Games are supposed to get more difficult as you progress through
from beginning to end (and merely surviving any given portion by the skin of
your teeth, using up all available resources, is supposed to make subsequent
parts even harder, so that as a player's skill progresses he gets better at the
parts of the game he has already seen and thus more able to face subsequent
portions, allowing him to make more and more progress), but NetHack gets very
much easier as the game progresses (and so many new NetHack players make very
little apparent progress in terms of the dungeon depth they usually reach until,
all of a sudden, they have a "lucky game" wherein they don't die in the early
game and in many cases go on to ascend the first time they ever get past about
quest depth; and then like as not their very next game they die on dungeon level
three or four, which seems to imply that skill is barely relevant at all, to how
far a player can get, although of course experienced players know that skill is
in fact highly relevant to how frequently a player can reach the end game;
nonetheless, the difficulty curve clearly needs work).

I think the dungeon design is indeed a large part of the difficulty curve. Most existing proposals favor overhauling Gehennom to make it harder and leaving the early dungeon relatively untouched, to bump up the difficulty of the late game. This proposal is no different, and that's a good thing.

In terms of Minetown's overall role in the dungeon as a whole, very little has
changed, apart from converting the areas outside the town proper from rooms to
caverns.  (Implementation-wise, this can be done by generating a normal cavern
map sans stairs, then overwriting the central portion of it with the embedded
town and placing the stairs in the areas outside the town.  Finally, each
section of outer town wall edged by a contiguous line of open floor gets exactly
one door.)

Doesn't this exist already? Maybe it was new in 3.6.

The Gnomish Sewer

I'd call this just The Sewer.

The entire level is dark.  The three spots marked ; are deep water and contain
electric eels; one of them also contains a (submerged) wand of cold, which can
be fetched in any of the usual ways for getting things out of water.  The
luckstone and the loadstone (one each) and the gemstones and glass (similar
amounts to the other Mine's End levels) are all positioned randomly, along with
several piles of coins, seven random rings, four random objects, a rotted pair
of high boots, three rust traps, six random traps, six sewer rats, three random
r, two shriekers, four random F, four random S, two crocodiles (or if they can
be called "aligators" so much the better), at least six and possibly as many as
nine random ; (positioned in water but otherwise at random), a 50% chance of a
chameleon, a 75% chance of a purple worm, and a 5% chance of a gremlin.  The up
stair can also be positioned randomly.  The random ; try to respect difficulty
level when the level is generated, to the extent that this is possible.

Some of the random gems and glass, and maybe some of the gold, can be embedded
in rock.  I'm tempted to say that some of the randomly-positioned objects
(especially the rings) can be submerged in random water tiles if the RNG happens
to plonk them there, but that's negotiable.  Flavor-wise it seems almost
mandatory, but I'm not sure whether it would be balanced from a gameplay
perspective, which is more important.

I would place one or two rings completely randomly and let the others be on dry land. It looks hard enough to navigate this level without chancing that the majority of the rings will be submerged.

If this could be made to generate the floor and water paths randomly it would be really cool.

Not sure I like the spaces the player has to squeeze or dig through to access, though I suppose this is the Mines and that should be expected.

I assume this level is diggable since it doesn't specify otherwise.

The Bottom of the Barrel

So is this supposed to be the "plain" version of Mines' End? Everything just lying around with no real gimmicks?

Not important, but I'd expect a level with this name to be circular.

Medusa's Island D ("Dances with Demiliches")

The walls are undiggable, but the floor is normal.  Medusa is standing on the
down stairs, surrounded by the random statues, the statue of Persues, two statue
traps, two water nymphs, and two cockatrices.  The entry rooms at the east and
west ends of the building each contain two centaurs and a demilich with a wand
of sleep, fire, cold, or lightning (equal chance of each type).  The long narrow
rooms to each side of Medusa's room each contain ten random monsters and five
almost-random traps, which cannot be trapdoors, pits, levelporters, or holes.
The water contains the usual collection of aquatic monsters (similar to the
other Medusa's Island levels), and the level also contains two dragons of random
color and several additional random monsters that can be placed anywhere.  The
statue of Perseus contains a shield of reflection, a sack, a blessed +2
scimitar, and a 50% chance of levitation boots.

Giving monsters attack wands in these new variants (also C, which gives them to storm giants) seems a bit counterproductive to the level's new status as a guaranteed source of reflection.

Doctor Jeckyl's Laboratory

The tiles marked # are sinks, and the ones marked \ are benches.  Benches are a
new type of furniture.  Any object sitting on a bench (if it's on the tile, it's
assumed to be on the bench) is out of reach for tiny monsters (except for ones
that can fly).  Sitting or resting on a bench provides regeneration (i.e., you
heal one movement point per turn), but this only works on turns when you #sit or
rest and only as long as you remain on the bench.  Of course, benches in the
Laboratory are presumed to be lab benches, and each of them has two or three
stacks of potions (see below for details).  Three of the benches also each
contain one tinning kit each.

If the potions are just sitting out in the open, won't the roaming monsters just collect them and drink down all the good ones before the player can get his hands on them?

Ken Arnold Memorial Library

The library building also contains four random L, two titans, two golden nagas,
four vampire mages, two barrow wights, three random "player character" monsters,
and six random traps, all positioned randomly.  Outside the library there are
four random dragons (or perhaps six).

Seems like the same problem on a smaller scale: won't they take and use some scrolls (I guess mainly teleportation, or possibly earth) the player would have liked to have? Of course, in FIQHack or Grunt, where monsters can read basically any scroll against you, this would be much worse.

Both of the spots marked ( have a chest.  One contains random chest contents.
The other is on burned Elbereth and contains a pair of lenses, 2-3 markers, an
athame, and possibly a scroll of wishing (see below).

The most flavor-appropriate way I can think of to include a wish here is to add
a Scroll of Wishing to the chest.  If included, it would be a non-stacking (so
you can't wish for more than one), non-randomly-generated (and not generated by
polymorph) random-appearance scroll (the new label could be RATSANOPU) that
cannot be written (or requires prohibitively many charges) and gives you one
wish when read if non-cursed and non-impaired.  If the scroll is cursed, you are
not prompted but still get a random wish item.  If you read the scroll while
confused, you mispronounce the magic words, and a random tame monster is
generated.  Whether it is worth adding an item to the game simply to add a wish
to this level is an open question, however.  It may be that the two magic
markers, combined with a significant collection of books and scrolls, can
adequately compensate for the lack of a wish.

RATSANOPU is a great scroll label in general.

Sample maps

Valley of the Dead with large graveyards

Given that graveyards contain a lot of chests which tend to contain a lot of spellbooks, and the graveyards here are fairly larger on average than vanilla, I don't think any more worry about the spellbook polymorph bug fix is needed.

Garden of Temptation

(I would like to also include sirens, who lure unwary
adventurers into the water to drown, but I am not sure it is worth adding a type
of monster and a type of attack to the game just for this level.  Opinions on
this question are wanted, but only if you can explain reasons; simple yes/no
votes are unhelpful.)

I can see other uses for sirens than this level: Medusa's Island, for one, could have a couple, as could any Gehennom level that generates water. Honestly, an attack which paralyzes and drags the player one square towards the siren has a lot of possible nasty uses.

Wizard's Tower, Bottom Level

---------------------
|...................|
|--S------------....|
|....|.}}}}}}}.|--S-|
|hive|.}}---}}.|....|
|....|.}--^--}.|....|
|-S--|.}|L<V|}.|....|
|....|.}--^--}.|...D|
|....S.}}---}}.|....|
|....|.}}}}}}}.|..^.|
---------------------

The path around the level, while still circuitous, has been shortened.  In order
to accommodate this, the beehive has been relocated to the west side of the
tower.  The random L is replaced with a guaranteed arch-lich or, if those are
genocided before the level is generated, a random &.  The vampire lord is the
same.  The two D (one marked, one placed randomly) are now great dragons of
random color.  The water contains two giant eels, two krakens, and two electric
eels.  (The electric eels are an addition.)  As before, there are also three
random demons, a random L, two random scrolls, a random weapon, a random potion,
a random tool, and a random amulet on the up ladder.

What's the point of the row of floor tiles on the far side of the moat?

Wizard's Tower, Middle Level

---------------------
|..<.|..............|
|-S--|-S----------S-|
|..|"|.........|....|
|..S.|.........|-S--|
|..|.|..demon..|....|
|S--S|...zoo...|....|
|....|.........|--S-|
|....S.........|....|
|....|.........|>...|
---------------------

Again, the path around the level is still circuitous but has been shortened.
Every tile in the marked demon zoo contains a random trap (not a hole or
trapdoor or level teleporter), a random hostile non-unique & set to ambush you,
a random wand, and either gold or a random non-glass gemstone.  As before, there
are two scrolls, two potions, and a spellbook positioned randomly within the
tower (but not in the zoo).

63 random wands? That's a LOT of loot... also means a wand of wishing might not be uncommon to find here. I calculated the probabilities for it out in another section.

A random trap on every square does sound like it'd slow the player down, though, and that seems like it goes against the impetus to make the tower less tedious (if you're getting caught in bear traps and setting off landmines all over the place, that will seriously slow down the game in real time.)

Wizard's Tower, Top Level

---------------------
|....S.......S......|
|....|-----------S--|
|....|}}}}}}}|......|
|....|}}---}}|......|
|....|}--^--}|......|
|--S-|}|d@V|}|----S-|
|....|}--^--}|......|
|.>..|}}---}}S......|
|....|}}}}}}}|......|
---------------------

The circuitous path has been shortened.  As before, the legendary Wizard stands
on the Book of the Dead, surrounded by four squeaky boards, two of which are
occupied by a hellhound and a vampire lord.  The moat contains four eels, four
krakens, and two piranhas.  On the floor there are six random demons, a random
dragon, two random giants, three random spellbooks, two random scrolls, two
random potions, a ruby, a spiked pit, a sleeping gas trap, an anti-magic trap,
and a magic trap.

It looks like the Tower itself is reduced in size? Not that I have a problem with this.

I agree that the vanilla path through the tower is unnecessarily winding, which is frustrating. Remove sources of frustration at all costs!

Sample Maze Level 01

Seems very biased towards vertical corridors, but maybe that's just because it appears to favor placing a | rather than a - at certain corners (i.e. (4,2)). I like the occasional loops.

Other Unsorted Notes

"Town" filler level generation algorithm: generates streets with rooms like the Tourist quest locate level.

Proposed new branch: The Mazes of Mystery - branch of Gehennom for those who actually like mazes. Generates several lit 3x3 rooms in each level, each containing a chest. Chest contents are biased towards scrolls and spellbooks, and may contain tools. Rooms might have multiple exits and might have secret doors generated where their doorway would be. Between the rooms is unlit maze filler, but it's not a perfect maze, some loops are allowed. Unlike standard Gehennom mazes, walls are undiggable. Levels are unmappable. Not sure whether the floor should be diggable. The stairs are placed a certain distance away from each other, perhaps in rooms.

I don't really see a "prize" for such a branch, so it doesn't dead-end like most branches. I see it as an optional path you can take to descend lower into Gehennom, like the Fire Pits and Swamp of Death. Of course, it needs to have some pretty dangerous things added to it to keep up with the increased difficulty of Gehennom.

Also can contain differently shaped and other types of rooms too, that contain things besides chests. Such as a statue of Theseus.

Lots of minotaurs are a given in this branch, it goes without saying.

How to allow loops: is it computationally feasible to calculate the farthest-away two "adjacent" points separated by a wall, break down this wall, and repeat N times? That would be a good way to do it.